fix: speed up dashboard loading

[numnx]

What changed

• Split noncritical dashboard shell features and modal flows out of the initial bundle.
• Added a compact active-sprint task projection for Overview and header counters while preserving full task records elsewhere.
• Removed duplicate Live/readiness reads and reduced execution-heartbeat refresh pressure on wide Overview resources.
• Corrected Tailwind CSS module references that duplicated generated framework CSS.
• Updated canonical and published dashboard documentation.

Why

The landing page eagerly loaded large optional chunks, fetched multi-megabyte read models it did not render, and repeatedly refreshed expensive readiness and analytics projections during active execution.

Performance verification

• Cold DOM-ready: ~5.4 s -> ~159 ms
• Cold first content: ~5.7 s -> ~432 ms
• Overview task payload: ~3.04 MB -> ~24 KB
• Initial JavaScript: ~1.39 MB -> ~567 KB
• Initial CSS: ~1.15 MB -> ~659 KB

Measurements were taken against the local production build on the dashboard loopback URL with a fresh headless browser context.

Validation

• pnpm run lint
• pnpm run test:backend — 5,743 passed, 7 skipped
• pnpm run test:dashboard — 2,279 passed, 1 skipped
• pnpm run build
• Cold-browser navigation, paint, long-task, and request trace against port 4444

Screenshots are not included because the change is performance-focused and does not intentionally alter dashboard appearance.

Risk and rollback

Risk is limited to Overview read-model freshness and lazy shell boundaries. Full task APIs remain unchanged by default, detailed Stats remains realtime, and all reduced-frequency resources retain polling or structure-event refreshes. Roll back this PR to restore the prior loading behavior.